Neighborhood-Level Socioeconomic Status and Prescription Fill Patterns Among Patients With Heart Failure

Key Points Question Is living in a neighborhood with low neighborhood-level socioeconomic status (nSES) associated with increased nonadherence to life-saving medications for heart failure? Findings In this cohort study of 6247 patients with heart failure with reduced ejection fraction, patients living in neighborhoods with lower nSES had significantly higher odds of nonadherence to guideline-directed medical therapy. Meaning These findings suggest that interventions to improve medication adherence among patients with heart failure should consider neighborhood-level approaches and policies to target patients at highest risk.


Introduction
Treatment for heart failure with reduced ejection fraction (HFrEF) includes guideline-directed medical therapy (GDMT) 1 with 4 categories: (1) β-blocker (BB); 2) angiotensin-converting enzymeinhibitor, angiotensin receptor blocker, or angiotensin receptor neprilysin inhibitor (ACEI/ARB/ARNI); 3) mineralocorticoid receptor antagonist (MRA); and 4) sodium-glucose cotransporter-2-inhibitor (SGLT2i).GDMT has improved clinical outcomes in multiple trials, 1 and contemporary analysis estimates a 4-fold decrease in mortality with this regimen. 2Despite substantial benefits, up to 40% to 50% of patients do not regularly fill these medications, [3][4][5] leading to increased hospitalization and mortality. 6ighborhood characteristics and the built environment (ie, the physical makeup of a community) may play a meaningful role in a patient's ability to fill medications.For example, a patient may have difficulty filling prescriptions if they have inadequate access to transportation, are unable to walk to the pharmacy, or do not have pharmacies nearby.Low neighborhood socioeconomic status (nSES) has been associated with medication nonadherence in other clinical contexts, such as diabetes, 7 chronic kidney disease, 8 and cardiovascular disease prevention. 9Patients with HFrEF may be especially vulnerable to these effects given the high complexity of HF medication regimens 10 and the frequent titration of GDMT doses and pills. 113][14][15] Because individual-level SES information may not always be readily available, identifying neighborhood-level risk factors for medication nonadherence could play a key role in informing community-based and policy-level interventions to improve medication adherence.Therefore, we aimed to assess the association between nSES and medication nonadherence to GDMT for patients with HFrEF.[18]

Study Design, Data Sources, and Participants
This retrospective cohort study included patients from a large health system in New York (NYU Langone Health).Data were obtained through the electronic health record (EHR) (Epic Systems), which was linked to medication fill data from pharmacies and pharmacy benefit managers nationwide (Surescripts) and census-tract level data (US Census Bureau's American Community Survey [ACS], 19 NYC Department of Transportation Transit Travelshed Model, 20 Walk Score ratings, 21 and National Neighborhood Data Archive National Establishment Time Series 22 ).Adults with the following were included: Patients with improved LVEF were included given recommendations to continue GDMT in this population. 23The study was approved by the NYU Grossman School of Medicine institutional review board and met criteria for a waiver of informed consent based on Federal Policy for the Protection of Human Subjects (45 CFR 46.116), including that the research involved no more than minimal risk to participants and could not be carried out practicably without the waiver.This report follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. 24

Measures Primary Exposure
To obtain nSES, we geocoded patient addresses (Geocodio) 25 for linkage to the US Census Bureau's ACS, 19 using ACS data from 2015 to 2019 for primary analyses to ensure the exposure was measured prior to mediators and outcomes for mechanistic interpretation.ACS variables were used to compute the Agency for Healthcare Research and Quality SES score, 26 which combined information on crowding, property value, unemployment, poverty level, income, and education.This score was categorized in quartiles, with the lowest quartile corresponding to lowest nSES.Patients with missing nSES (ie, due to PO box address or missing ACS variables) were included in models as a separate category to limit selection bias. 27

Outcomes
Nonadherence to GDMT over 6 months was measured using the proportion of days covered (PDC) metric recommended by the National Quality Forum and Pharmacy Quality Alliance (PQA). 28,29As previously described by others, 30 PDC was defined as the ratio of days a medication was filled to days a prescription was active; lapses in prescription were excluded from the denominator.Two data sources were used to compute PDC: (1) pharmacy fill information (Surescripts) and (2) prescription information (EHR).As previously described by our group, 31 this EHR-linked data set was comprehensive and contained more prescription fill information than insurance claims data.PDC was measured over 6 months from time of prescription order and averaged across GDMT categories if more than 1 GDMT was prescribed, as previously described. 32Patients who died during the study were censored.Given our aim to study chronic GDMT, medication categories prescribed for fewer than 28 days were excluded from calculations.The primary outcome was binary: mean PDC less than 0.8 (nonadherent) and mean PDC greater than or equal to 0.8 (adherent).This classification is widely used in the literature, recommended by the PQA, and has previously been associated with rehospitalization and death among patients with HFrEF. 29,33PDC was also tabulated as a continuous outcome for descriptive purposes.

Neighborhood-Level Mediators
The following neighborhood-level mediators were assessed: access to transportation, walkability, and pharmacy density (direct acyclic graph in Figure 1 assumes causal relationships have no feedback).Access to transportation was measured using the 2018 NYC Transit Travelshed Model, 20 which computes area reached within 60 minutes by walking, rail, subway, bus, and ferry.Walkability was assessed using the 2022 Walk Score, 21 which rates ease of walking from 0 (lowest) to 100 (highest).Pharmacy density was obtained from the 2017 National Neighborhood Data Archive 22 and was defined as number of pharmacies per 10 000 people.

Statistical Analysis
Baseline characteristics were assessed across nSES categories with χ 2 tests or analysis of variance (ANOVA) as appropriate.Multivariable logistic regression models were used to assess the association between nSES categories and odds of nonadherence, with and without adjustment for age, sex, most recent LVEF, comorbidity index, hospitalization in the past year, and ED visit in the past year.For interpretation, difference in probability of nonadherence between each nSES category and the highest quartile, averaged over covariates, was derived from the fitted logistic regression model and reported as a measure of total effect.We chose not to include individual patient race, ethnicity, language, and insurance status in multivariable models.This is because current and historical institutional and structural racism has led to racial and ethnic segregation among neighborhoods, and variations in health behavior by race, ethnicity, or language may, in part, be mediated by nSES factors. 35,36Moreover, these variables may be associated with individual-level SES factors, such as income, wealth, and education, which were not reliably available in our data set.Furthermore, associations between individual-level and neighborhood-level SES variables could be bidirectional and lead to a combination of mediation and/or confounding mechanisms. 37Inclusion of these variables without being able to adjust for these other associations could lead to biased or incomplete conclusions.Therefore, our primary analytic model aimed to assess nSES as a global risk factor for nonadherence, rather than attempting to isolate the independent effect of nSES from the myriad of measurable and unmeasurable individual-level factors.
In secondary stratified analyses, to evaluate for the specific intersectionality of nSES and race and ethnicity, we assessed interactions between race, ethnicity, and nSES.To do so, we used stratified multivariate logistic regression models similar to the primary analysis for the following subgroups: race (Black vs White) and ethnicity (Hispanic/Latinx vs non-Hispanic/Latinx). ANOVA test was used to assess P for interaction for adjusted models with and without interaction term for each subgroup.
Separate mediation analyses were conducted to estimate the extent to which nSES disparities in nonadherence were mediated through each built environment mediator.For each mediator, multivariable logistic regression models were used to assess the associations between (1) nSES and built environment factor (association of exposure with mediator) and (2) nSES and nonadherence (association of exposure with outcome), adjusting for the same covariates described previously.
Estimated coefficients from the 2 models were combined to estimate natural average direct and natural average indirect effects as differences in probabilities of nonadherence.Bootstrapped confidence intervals with 1000 simulations were computed. 38Proportion mediated was reported when the indirect effect was significant and along the direction of the direct effect.Due to data availability, analysis for transportation was restricted to patients in New York City.Additionally, analysis for pharmacy density used 2013-2017 ACS data to allow for exposure prior to mediator.
Analyses were conducted using R version 4.2.2 (R Project for Statistical Computing).Two-sided P < .05indicated statistical significance.

Baseline Characteristics
Among 6247 included patients, the mean (SD) age was 73 ( 14) years, and majority were male (4340 [69.5%]).There were 1011 (16.2%)Black participants, 735 (11.8%)Hispanic/Latinx participants, and 3929 (62.9%) White participants.Overall, 599 (9.6%) had missing nSES data, and remaining patients were evenly distributed across nSES quartiles, with 3555 (55.3%) living in New York City.Several baseline characteristics varied by nSES category (Table 1).Patients living in lower nSES areas tended a P values reflect unadjusted differences across the 5 neighborhood socioeconomic status categories using χ 2 tests for proportions or analysis of variance for means as appropriate.
b Other race included American Indian or Alaska Native, Asian, and Native Hawaiian or Other Pacific Islander.
c Patients were required at least one LVEFՅ40 for inclusion, but were still included if LVEF recovered.
d Access to transportation was measured using the transportation mobility index, defined as the geographical area in acres able to be reached within 60 minutes of transit.
e Walkability was measured using the Walk Score rating on a scale of 0 to 100, with higher score indicating greater ease of walking within a neighborhood.
f Pharmacy density was measured as number of pharmacies per 10 000 people.
to be younger; were more often Black or other race, Hispanic/Latinx ethnicity, Spanish-speaking; had Medicaid insurance, lower LVEF, and lower comorbidity index scores; more often visited the ED in the past year; and were more often prescribed ARNI.Access to transportation and pharmacy density were lower for patients in lower nSES areas, and walkability was higher.

Association Between nSES and Medication Nonadherence
The percentage of patients with nonadherence to GDMT (PDC <0.

Interaction and Mediation Analyses
There was no significant interaction between race or ethnicity on the association between nSES and odds of nonadherence (eFigure in Supplement 1).There was no significant indirect effect (ie, effect through the mediator) observed for access to transportation or pharmacy density (Table 3).For walkability, a small, but statistically significant indirect effect was observed, such that walkability contributed to approximately 7% of the variability in probability of nonadherence for patients in nSES quartiles 1 or 2 compared with quartile 4 (percentage mediated, P < .001).In absolute terms, higher walkability appeared to be protective in quartiles 1 and 2 compared with quartile 4 and contributed to an estimated improvement in nonadherence by 0.7% and 0.5%, respectively (indirect effects, P < .001).

Discussion
In this retrospective cohort study of patients with HFrEF, living in a lower nSES area was associated with higher odds of nonadherence to GDMT as measured by pharmacy fills.This association persisted after adjustment for clinical covariates.Our findings highlight an important inequity in HFrEF treatment, reveal a key risk factor for medication nonadherence, and underscore the  3) g 0.5 (0.2 to 0.9) g 7.2 (3.8 to 10.9) g 6.9 g 1 10.4 (6.6 to 14.1) g 0.7 (0.3 to 1.2) g 11.1 (7.5 to 14.7) g 6.5 g Pharmacy density

[Reference] [Reference]
[Reference] NA b Natural direct effect is the estimated difference in rates of nonadherence independent of the built environment factor for patients in 1 nSES quartile when compared with the highest nSES quartile, ie, the difference in risk if the built environment factor was the same in both nSES quartiles.
c Natural indirect effect is the estimated difference in rates of nonadherence transmitted through the built environment factor for patients in 1 nSES quartile compared with the highest nSES quartile, ie, the difference in risk that is changing due to changes in the built environment factor.
d Total effect is the sum of the direct and indirect effects.Note, total effects are different for the different mediation analyses due to differences in cohort (New York City-only cohort for access to transportation) and exposure (2013-2017 nSES calculation for pharmacy density).
importance of considering neighborhood-level disparities when developing initiatives to improve medication adherence.
Patients with HFrEF are known to have considerable shortfalls in medication adherence that contribute to increased mortality and hospitalization. 6Our findings inform the hypothesis that neighborhood-level disparities contribute to these shortfalls.Overall, nonadherence to GDMT was common in our cohort, occurring in 52% of patients in the lowest nSES quartile and 40% of patients in the highest nSES quartile.This range of nonadherence is consistent with other studies in HF, [3][4][5][6]39 further confirming the extent of this problem. Morer, the magnitude of the disparity in nonadherence by nSES that we observed (risk difference >10%) was likely clinically meaningful, given that decreases in GDMT adherence by even 5% to 10% are associated with significant increases in mortality among patients with HF. 6,39 Given that patients with HFrEF living in lower nSES areas tend to have increased mortality and readmission, 40 understanding and mitigating barriers to medication adherence in this population has the potential to ultimately reduce inequities in clinical outcomes.
9]41 Our study also addressed limitations of prior work that utilized administrative claims data, 8,9 which are often inherently restricted to a single insurance type; lack the ability to incorporate clinical covariates, such as LVEF; and are unable to account for prescribing actions.
Overall, the association between nSES and medication nonadherence is likely multifaceted and due to complex, dynamic interactions between environmental, individual, and cognitive factors. 42ighborhood-level factors include environmental stressors, such as violence and perceived safety, which could serve as physical barriers 43 to obtaining medications, and also act as emotional stressors that drain one's capacity for adherence in the setting of immediate threat. 12Additionally, built environment factors, such as access to transportation, walkability, and access to health services, could also impede one's ability to physically obtain medications, or could impede interactions with health care professionals to discuss medication-related concerns. 44Moreover, neighborhood-level social factors contributing to medication nonadherence could include social cohesion and social norms about medications. 12,45Beyond neighborhood-level factors, associated individual-level factors could include medication cost, income, family support, health literacy, health care interactions, language, and individual beliefs. 12For example, several studies in HF have shown an association between income, 15 medication costs, 14,[46][47][48] and nonadherence to GDMT, most recently for the newer classes, ARNI 14,15 and SGLT2i. 13In our study, patients living in lower nSES areas may have had lower income, which could make them more vulnerable to medication nonadherence.
In our mediation analysis, we had hypothesized that built environment barriers to physically obtaining medications would partly explain neighborhood-level disparities in nonadherence.
Surprisingly, we found no mediation by access to transportation or pharmacy density, and only a small mediation effect by walkability.This suggests that the association between nSES and medication nonadherence is more likely due to the myriad of other environmental and individual factors contributing to medication nonadherence discussed previously.It is also likely that the built environment measures that we used did not capture the complex manner in which patients obtain medications.For example, disparities in pharmacy accessibility can go beyond location, and can include disparities in hours of operation, availability and price of medications, engagement with pharmacists, and discrimination by staff.

Limitations
This was a retrospective cohort study conducted across a single health system in an urban environment.PDC measures medication adherence through pharmacy fills, which may not reflect what the patient is actually taking.We chose a PDC threshold of less than 0.8 based on previously described associations with adverse outcomes in HFrEF, 33 but the optimal PDC threshold for this population is unknown. 58We did not assess whether prescriptions were filled by mail.Due to the nature of EHR data, we were unable to reliably account for time since HF diagnosis or HF treatment history.The relationship between neighborhood-level and individual-level factors is complex, 37 and due to our limited EHR data on individual-level SES variables, we chose to assess nSES as a global risk factor.P values for interaction and mediation analyses were not adjusted for multiple testing and should be interpreted cautiously.Of note, we found no statistically significant interaction between individual race or ethnicity on the association between nSES and nonadherence, which may have been due to the relatively small proportion of Black and Hispanic/Latinx patients in our sample.
Although we accounted for temporality by using primary exposure data that was collected prior to mediators and outcomes, we could not account for neighborhoods changing over time, patients moving between neighborhoods, or patients leaving the area.Walkability data were obtained 1 year after the study period, although it is unlikely to have changed substantially over a year.We included patients with missing nSES as a separate exposure category to reduce selection bias, as previously described. 27Interpretation of associations between missing nSES and nonadherence is limited due to incompletely understood reasons for missing data.While our approach was based on a directed acyclic graph guided by the theoretical framework to assess mechanistic associations, important confounders may be unmeasured or missing.

Conclusions
We found higher rates of nonadherence to GDMT among patients with HFrEF living in lower nSES areas.This association remained after adjustment for clinical characteristics.

Figure 1 .
Figure 1.Direct Acyclic Graph Depicting Mediation Analyses Abbreviations: NA, not applicable; nSES, neighborhood-level socioeconomic status.aAdjusted for age, sex, ejection fraction, emergency department visit in past year, hospitalization in past year, and Elixhauser comorbidity index.

Table 1 .
Baseline Characteristics by Neighborhood-Level Socioeconomic Status

Table 2 .
Unadjusted Rates of Medication Nonadherence by Neighborhood-Level Socioeconomic Status

Table 3 .
Estimated Differences in Probabilities of Nonadherence for Direct, Indirect, and Total Effects of the Association Between nSES and Medication Nonadherence and Percentage Mediated by Built Environment Factors a 573][54][55][56]ed interventions have been shown to improve medication adherence and reduce mortality and hospitalization for patients with HF.52These often involve patient education, monitoring, and frequent follow-up with health professionals.Our findings support the adaptation of these interventions for patients living in at-risk neighborhoods.For example, involving community health workers and local pharmacists in such programs could help patients overcome distrust, link patients to community resources, and provide up-front face-to-face counseling.[53][54][55][56]Additionally,screening for neighborhood-level risk in clinical practice57could provide opportunities to match patients to local programs.
order services could mitigate built environment barriers, and we did not assess this.Therefore, although increasing pharmacies, adding transportation routes, and improving walkability likely has benefits, these efforts are probably insufficient to address nSES disparities in medication nonadherence.
Our findings highlight an important gap that may contribute to known neighborhood-level disparities in HF care and outcomes.Future investigations and initiatives should focus on identifying neighborhood-level barriers to adherence, targeting interventions to neighborhoods at highest risk, and adapting them to address community needs.This is an open access article distributed under the terms of the CC-BY License.© 2023 Mukhopadhyay A et al.JAMA Network Open.Amrita Mukhopadhyay, MD, Division of Cardiology, Department of Medicine, NYU Grossman School of Medicine, 227 E 30th St, 8th Floor, New York, NY 10016 (amrita.mukhopadhyay@nyulangone.org).